Key Info
Tencent Hunyuan's Hy4 has been ranked the #1 Chinese model on Agent Arena, based on 12K+ real-world long-horizon agentic sessions, and scores 35 on AA Intelligence Index v5, placing second among open models.
Highlights
- Hy4 shows a +5.4% net improvement and sits on the Agent Arena Pareto frontier at a $0.57 median cost per task.
- Tencent Hunyuan is the only Chinese lab in Agent Arena's top 5.
- On AA Intelligence Index v5, Hy4 scores 35, tied with Grok 4.7 and behind only Kimi K3.1 Pro (46) among open models.
- Hy4 leads Chinese models in agentic coding with 31% on FrontierCode 2.0 (DeepSeek 28%, Kimi 25%).
- Cost per task is reported at $0.78, with WorkBuddy expanding inference capacity 10x after launch-day demand and extending free Hy3 access to Nov 15.